Practical safety in flexible access control models

Abstract

Assurance that an access control configuration will not result in the leakage of a right to an unauthorized principal, called safety , is fundamental to ensuring that the most basic of access control policies can be enforced. It has been proven that the safety of an access control configuration cannot be decided for a general access control model, such as Lampson's access matrix, so safety is achieved either through the use of limited access control models or the verification of safety via constraints. Currently, almost all safety critical systems use limited access control models, such as Bell--LaPadula or Domain and Type Enforcement, because constraint expression languages are far too complex for typical administrators to use properly. However, researchers have identified that most constraints belong to one of a few basic types, so our goal is to develop a constraint expression model in which these constraints can be expressed in a straightforward way and extensions can be made to add other constraints, if desired. Our approach to expressing constraints has the following properties: (1) an access control policy is expressed using a graphical model in which the nodes represent sets (e.g., of subjects, objects, etc.) and the edges represent binary relationships on those sets and (2) constraints are expressed using a few, simple set operators on graph nodes. The basic graphical model is very simple, and we extend this model only as necessary to satisfy the identified constraint types. Since the basic graphical model is also general, further extension to support other constraints is possible, but such extensions should be made with caution as each increases the complexity of the model. Our hope is that by keeping the complexity of constraint expression in check, flexible access control models, such as role-based access control, may also be used for expressing access control policy for safety-critical systems.

Cite this work

No public copy is stored for this record, so only the metadata is cited.

Stored only when both papers are in the corpus — 116 citations exist upstream.

Papers that cite the same references — the relational fallback while embeddings are still backfilling.